Driver and traveler behaviors constitute the core of traffic simulation models. The reliability of the simulation tool depends on the realism of the underlying behavior models. Understanding these behaviors is also important for traffic safety studies and development and evaluation of driver assistance systems. The development of these models involves theoretical formulation of the model structure equations based on proven behavioral principles and estimation of the parameters of the model using advanced econometric methods, such as latent variables and classes to represent unobserved driver characteristics (e.g. aggressiveness). We have studied the behaviors of acceleration, lane choice and lane changing and overtaking, as well as route choice and route switching.
Development of driver and traveler behavior models requires detailed data on the movement of subjects in the traffic stream and their interactions with other vehicles. Increasingly, sophisticated data collection techniques, which involve GPS and cellular transmission technologies, and automated data reduction methods that apply image and pattern recognition algorithms are being developed and used. In addition to real-world data collection we employ driving simulators, which consist of a vehicle located in a computer-controlled virtual reality laboratory that displays the outside “world” including the built environment, road facilities, signals and signs and surrounding vehicles. Within this laboratory environment, experiments and scenarios to collect hard-to-get data can be devised and the data gathered this way can be used in model estimation together with the available field data.
In the context of route choices we have focused on developing efficient algorithms to solve the problem traffic assignment problem where route choices are based on the Cross Nested Logit (CNL) model. This model accounts for the correlations among the utilities of overlapping routes. As a result, the model can better predict the selection probabilities of the various routes.